A Study on the Improvement of Life Insurance Underwriting using the Feature Selection Method and Ensemble Classification Model

ABSTRACT

With the changing workplace landscape evident from the recent remote working arrangement, owing to the disruption caused by COVID-19 pandemic, the pace of adopting integrated services with AI has accelerated. In the insurance industry, there has been a gradual increase in business cases and research, with the introduction of AI technology in areas such as detection of unfair claims, claims adjusting, and insurance acceptance. In this study, an insurance underwriting model for accepting/rejecting new applicants was developed to reduce these discrepancies, based on underwriters, as well as enable faster processing. The data of Prudential Life Insurance from Kaggle was utilized to develop the insurance underwriting model. Among the feature selection methods, the filter-based and embedded methods were comparatively evaluated, and a Regularized Random Forest from the embedded methods was finally selected. For the insurance underwriting model, seven classification algorithms were applied for model optimization, and using the ensemble voting, the result of models with excellent classification performance with a recall score of 0.8 or higher was finally predicted by voting to ensure derivation of reliable results.